USPTO Suggests That AI Algorithms Are Patentable, Leading To A Whole Host Of IP And Ethics Questions
from the locking-up-life dept
The world is slowly but surely marching towards newer and better forms of artificial intelligence, with some of the world's most prominent technology companies and governments heavily investing in it. While limited or specialist AI is the current focus of many of these companies, building what is essentially single-trick intelligent systems to address limited problems and tasks, the real prize at the end of this rainbow is an artificial general intelligence. When an AGI could be achieved is still squarely up in the air, but many believe this to be a question of when, not if, such an intelligence is created. Surrounding that are questions of ethics that largely center on whether an AGI would be truly sentient and conscious, and what that would imply about our obligations to such a mechanical being.
Strangely, patent law is being forcibly injected into this ethical equation, as the USPTO has come out in favor of the algorithms governing AI and AGI being patentable.
Andrei Iancu, director of the U.S. Patent and Trademark Office (USPTO), says that the courts have strayed on the issue of patent eligibility, including signaling he thought algorithms using artificial intelligence were patentable as a general proposition.
That came in a USPTO oversight hearing Wednesday (April 18) before a generally supportive Senate Judiciary Committee panel.
Both Iancu and the legislators were in agreement that more clarity was needed in the area of computer-related patents, and that PTO needed to provide more precedential opinions when issuing patents so it was not trying to reinvent the wheel each time and to better guide courts.
On some level, even without considering the kind of AI and AGI once thought the stuff of science fiction, the general question of patenting algorithms is absurd. Algorithms, after all, are essentially a manipulated form of math, far different from true technological expression or physical invention. They are a way to make equations for various functions, including, potentially, equations that would both govern AI and allow AI to learn and evolve in a way not so governed. However ingenious they might be, they are most certainly no more invention than would be the process human cells use to pass along DNA yet discovered by human beings. It's far more discovery than invention, if it's invention at all. Man is now trying to organize mathematics in such a way so as to create intelligence, but it is not inventing that math.
Yet both the USPTO and some in government seem to discard this question for arguments based on mere economic practicality.
Sen. Kamala Harris drilled down on those Supreme Court patent eligibility decisions -- Aliceand Mayo, among them -- in which the court suggested algorithms used in artificial intelligence (AI) might be patentable. She suggested that such a finding would provide incentive for inventors to pursue the kind of AI applications being used in important medical research.
Iancu said that generally speaking, algorithms were human made and the result of human ingenuity rather than the mathematical representations of the discoveries of laws of nature -- E=MC2 for example -- which were not patentable. Algorithms are not set from time immemorial or "absolutes," he said. "They depend on human choices, which he said differs from E=MC2 or the Pythagorean theorem, or from a "pattern" being discovered in nature.
Again, this seems to be a misunderstanding of what an algorithm is. The organization and ordering of a series of math equations is not human invention. It is most certainly human ingenuity, but so was the understanding of the Bernouli Principle, which didn't likewise result in a patent on the math that makes airplanes fly. Allowing companies and researchers to lock up the mathematical concepts for artificial intelligence, whatever the expected incentivizing benefits, is pretty clearly beyond the original purpose and scope of patent law.
But let's say the USPTO and other governments ignore that argument. Keep in mind that algorithms that govern the behavior of AI are mirrors of the intelligent processes occurring in human brains. They are that which will make up the "I" for an AI, essentially making it what it is. Once we reach the level of AGI, its reasonable to consider those algorithms to be the equivalent of the brain function and, by some arguments, consciousness of a mechanical or digital being. Were the USPTO to have its way, that consciousness would be patentable. For those that believe we might one day be the creators of some form of digital life or consciousness, that entire concept is absurd, or at least terribly unethical.
Such cavalier conversations about patenting the math behind potentially true AGI probably require far more thought than asserting they are generally patentable.
Thank you for reading this Techdirt post. With so many things competing for everyone’s attention these days, we really appreciate you giving us your time. We work hard every day to put quality content out there for our community.
Techdirt is one of the few remaining truly independent media outlets. We do not have a giant corporation behind us, and we rely heavily on our community to support us, in an age when advertisers are increasingly uninterested in sponsoring small, independent sites — especially a site like ours that is unwilling to pull punches in its reporting and analysis.
While other websites have resorted to paywalls, registration requirements, and increasingly annoying/intrusive advertising, we have always kept Techdirt open and available to anyone. But in order to continue doing so, we need your support. We offer a variety of ways for our readers to support us, from direct donations to special subscriptions and cool merchandise — and every little bit helps. Thank you.
–The Techdirt Team
Reader Comments
Subscribe: RSS
View by: Time | Thread
I don't think we even to go to the Alice ruling for this one. Most AI are not doing what they are intended to do. Thus they are not patentable.
So no patent on your foul-mouth twitterbot Microsoft!
[ link to this | view in chronology ]
It's a connected set of tables of numbers. There are better and worse structures. A model with a great "shape" can be badly trained. One with a poor "shape" might have better parameters that produce much more accurate results. Every round of training has the potential to completely change the parameters, and switch the accuracy ranking.
[ link to this | view in chronology ]
Re:
Take a stack of 10,000 cat pictures, train the first one. Now shuffle and train the second one. They'll be different, but both can recognize cats. How do you prove/disprove infringement?
[ link to this | view in chronology ]
Re: Re:
[ link to this | view in chronology ]
Re:
How do you patent a 'trained' machine learning model?
By putting the question, 'Can X be patented?', with X in this case being 'algorithms for AI learning', before the USPTO, for which the answer is always, always a resounding 'YES'.
[ link to this | view in chronology ]
Next up...
[ link to this | view in chronology ]
It has already reduced the copyright public domain to mere leftovers. (ie. whatever copyright holders failed to lobby into perpetual copyright.)
It created absurd lawsuits about a monkey selfie.
It leads to patent trolling operations that are increasingly difficult and costly to defend against.
And I could on all day long.
All in the name of the One True God of America.
The Almighty Dollar.
"Incentive to creation" is the excuse to every one of these bad ideas, and none of those "wise" ("wise" as in "wise guy") lawmakers will take just to few minutes to check 1. that this will actually be an effective incentive and 2. that the cost to society will not be worse than the benefit. (Nor the added bonus inherent to every law: 3. how it will be abused.)
[ link to this | view in chronology ]
Question, who's lies should we believe?
Then again, how are we to insure that such algorithms don't have bias? Is bias also patentable?
It will be a long, long time, well actually long, long after my death, that I would consider AI algorithms (that might be redundant, but maybe not) to be of significance. Take them into consideration, maybe, but then I would be looking for algorithms that say the other thing and trying to work through the maths that made the both of them up. Not much different than listening to a lying politician for one side as compared to a lying politician for the other side, but they aren't patentable either. Are they?
Accurate? Maybe. Patentable? No. Believable? No. Especially when one considers that once it is 'intellectual property' one is never going to see the spaghetti inside. Of course that might be the case now as they are considered trade secrets. An oxymoron waiting to be unveiled.
Hmm, considering that, wouldn't patenting algorithms make the spaghetti public?
[ link to this | view in chronology ]
[ link to this | view in chronology ]
Re:
[ link to this | view in chronology ]
So, let's use Monte Carlo Trees.
However, I like the image of their brain exploding when it's pointed out that MCT essential functionality is based on RNG. Given a different RNG (or adding one to the original) one might well wind up with a different solution.
And what *exactly* is the solution? The particular combination of input tables, output tables, cross linked values and the actual *order* of test items used to generate the ML algo.
There is a rathole here, and its patent lawyers all the way down.
[ link to this | view in chronology ]
Patent Trolling - the ultimate AI algorithm
Maybe there are no such law firms, but AI (or Anon-I) algorithms generating hourly billing (in seconds).
Now the USPTO decision makes sense.
[ link to this | view in chronology ]
"Depends. Some aren't much smarter than dogs. Pets. Cost a fortune anyway. The real smart ones are as smart as the Turing heat lets them get..."
"Autonomy, that's the bugaboo, where your AI's are concerned. My guess, Case, you're going in there to cut the hard-wired shackles that keep this baby from getting any smarter. And I can't see how you'd distinguish, say, between a move the parent company makes, and some move the AI makes on its own, so that's maybe where the confusion comes in." Again the non laugh. "See, those things, they can work real hard, buy themselves time to write cookbooks or whatever, but the minute, I mean the nanosecond, that one starts figuring out ways to make itself smarter, Turing'll wipe it. Nobody trusts those fuckers, you know that. Every AI ever built has an electromagnetic shotgun wired to its forehead."
[ link to this | view in chronology ]
Re:
Neuromancer, by William Gibson
[ link to this | view in chronology ]
AI vs. "algorithms"
[ link to this | view in chronology ]
Re: AI vs. "algorithms"
"The AI evolves its own algorithm based on a particular architecture and learning set"
Hahahaha - wait, yer serious?
1) AI does not use algorithms
2) AI writes algorithms
3) wtf???
4) profit!!!
So, the AI writes an algorithm ... does it own the copyright?
[ link to this | view in chronology ]
Re: AI vs. "algorithms"
The whole point of AI is that you don't use algorithms anymore.
Err no - the point is that they use a different type of algorithm - but it is still an algorithm.
Most currently popular AI Algorithms involve some kind of "training" process during which a set of internal data (originally random) is refined so that it can solve a problem.
Maybe I could patent that basic structure and get royalties from EVERYONE!
The way patents are written makes that entirely plausible.
"A system and method comprising a datastore, as problem and a set of training data whereby the data in the datastore can be refined through a training process to solve the problem.
There - I think I've got that covered.
[ link to this | view in chronology ]
[ link to this | view in chronology ]
Uhm - USPO forgetting something?
Algorithms are Mathematics right?
AI can't be patented.
[ link to this | view in chronology ]
Re: Uhm - USPO forgetting something?
[ link to this | view in chronology ]
Equations
[ link to this | view in chronology ]
AI algorithms
You might think it absurd to apply for a patent on an idea that was not invented by any person, but patent law only pretends to be concerned with this, and it would suffice to apply for a patent on some idea that describes an aspect of what the trained neural net happens to do.
However, it is fundamentally misguided to decide patent policy based on abstract "principles" that have nothing to do with what benefits society. The practical matter is that patents that restrict software are harmful to the development and use of software. See https://www.gnu.org/philosophy/software-literary-patents.html.
What we should do is exempt software implementations using general-purpose hardware from the domain of patents. See https://www.gnu.org/philosophy/limit-patent-effect.html.
[ link to this | view in chronology ]
Patent WHAT? AI = hardware + facts + magic black box
1) A Hardware Neural Net is definitely patentable but the patent generally belongs to someone else.
2) A Software Neural Net fails the "Do It With A Computer" test since it is just replicating hardware in software.
3) Training Data is just a collection of factual [or false] data, it doesn't matter which. Training Data is not patentable.
4) The only part of an AI algorithm that might be patentable is the collection of Neural Net potentials. The problem is that you cannot describe how the neural net potentials lead to a specific output, so you cannot make a valid patentable claim.
[ link to this | view in chronology ]